Non-Gaussian velocity distributions integrated over space, time, and scales.

Velocity distributions are an enhanced representation of image velocity containing more velocity information than velocity vectors. In particular, non-Gaussian velocity distributions allow for the representation of ambiguous motion information caused by the aperture problem or multiple motions at motion boundaries. To resolve motion ambiguities, discrete non-Gaussian velocity distributions are suggested, which are integrated over space, time, and scales using a joint Bayesian prediction and refinement approach. This leads to a hierarchical velocity-distribution representation from which robust velocity estimates for both slow and high speeds as well as statistical confidence measures rating the velocity estimates can be computed.     

Human behavior classification by analyzing periodic motions

Recognizing human action is a critical step in many computer vision applications. In this paper, the problem of human behavior classification is addressed from a periodic motion analysis viewpoint. Our approach uses human silhouettes as motion features that can be obtained efficiently, and then projected it into a lower dimensional space where matching is performed. After a periodic analysis, each action unit is represented as a closed loop in this lower dimensional space, and matching is done by computing the distances among these loops. The main contributions are twofold: (1) an efficient periodic action feature constructing method is introduced; and (2) the difference between action units with different phase is computed adaptively with a novel distance proposed in this work. To demonstrate the effectiveness of this approach, human behavior classification experiments were performed on an open dataset. Classification results are highly accurate and show that this approach is promising and efficient.     

Human behavior classification by analyzing periodic motions

Recognizing human action is a critical step in many computer vision applications. In this paper, the problem of human behavior classification is addressed from a periodic motion analysis viewpoint. Our approach uses human silhouettes as motion features that can be obtained efficiently, and then projected it into a lower dimensional space where matching is performed. After a periodic analysis, each action unit is represented as a closed loop in this lower dimensional space, and matching is done by computing the distances among these loops. The main contributions are twofold: (1) an efficient periodic action feature constructing method is introduced; and (2) the difference between action units with different phase is computed adaptively with a novel distance proposed in this work. To demonstrate the effectiveness of this approach, human behavior classification experiments were performed on an open dataset. Classification results are highly accurate and show that this approach is promising and efficient.     

Weakly and strongly non-minimum phase systems: properties and limitations

In this paper, a new classification of non-minimum phase systems based on stabilisability of the internal dynamics is introduced. According to this classification, the non-minimum phase systems are categorised as follows: (1) weakly non-minimum systems, for which a finite value of minimum norm of the regulation error can be obtained, and (2) strongly non-minimum phase systems for which this minimum norm of the regulation becomes infinite due to the presence of both transient and steady-state errors while for the weakly non-minimum phase systems there is no steady-state regulation error. It is worth mentioning that based on this new classification a level of non-minimum phase-ness can be further defined for any given desired trajectory. Furthermore, some numerical simulations are carried out to show consistency with theoretical analysis.     

Robust output tracking control of a class of non-minimum phase systems and application to VTOL aircraft

In this article, we study the output tracking control of a class of MIMO nonlinear non-minimum phase systems in the presence of input disturbances. In order to attenuate the effects of disturbances, the method of uncertainty and disturbance estimator (UDE) is extended to the controller design for non-minimum phase systems. Due to the fact that the accumulated disturbances is composed of internal states and external disturbances, a different stability analysis is given, and the overall closed-loop system is proved to be semi-globally stable. The proposed state-feedback controller not only forces system outputs to asymptotically track desired trajectories, but also drives the unstable internal dynamics to follow bounded and causal ideal internal dynamics (IID) solved via stable system centre (SSC) method. Simulation results demonstrate that the proposed controller achieves excellent tracking and disturbance rejection performance via the example of VTOL aircraft which has been the benchmark of nonlinear non-minimum phase systems.     

An Algebraic Formalism for Computing the Moments of Distributions of Quadratic Forms

Distribution moments for quadratic forms are computed. Formulas for moments are derived with the use of the algebra of symmetric polynomials and invariants. The results are applied to parametrization and polynomial approximation of distributions in non-Gaussian statistical models of multialternative classification.     

Inverse Dyamics of Non-Minimum Phase Systems with Non-Zero Initial Conditions

A method is presented forcomputing the inverse dynamics of a linear non-minimum phasesystem with non-zero initial conditions. The method is also usedto change or correct a trajectory after it is already in motion,and consequently, it will allow for real time control by continuallyupdating the inverse dynamics computation.Frequency domain techniques are used to compute the input functionneeded to produce a desired output trajectory at a particulardegree of freedom. An output profile based on the differencebetween the desired trajectory and either a homogeneous responseor a forced response to a previous forcing function is used tocompute the required input function. The resulting input functionactively damps out initial conditions in the system and makesit track the desired trajectory.The method is applied to a non-collocated single-link flexiblerobot arm. The finite element method using Timoshenko beam theoryis used to discretize the equations of motion. Torque profilesare computed to control the tip displacement for several problems.The first problem is to control the tip to a desired trajectorywhen starting with non-zero initial conditions. The second problemis to change the desired trajectory while the previous desiredtrajectory is already in motion. The third problem is to correctthe trajectory after a disturbance is added to the system. Thefourth problem is to analyze sensitivity to errors in the modeland initial conditions. The last problem is to compare tip responsesfor rigid and flexible link assumptions in the inverse dynamicscomputation.     

A method for improving classification reliability of multilayerperceptrons

Criteria for evaluating the classification reliability of a neural classifier and for accordingly making a reject option are proposed. Such an option, implemented by means of two rules which can be applied independently of topology, size, and training algorithms of the neural classifier, allows one to improve the classification reliability. It is assumed that a performance function P is defined which, taking into account the requirements of the particular application, evaluates the quality of the classification in terms of recognition, misclassification, and reject rates. Under this assumption the optimal reject threshold value, determining the best trade-off between reject rate and misclassification rate, is the one for which the function P reaches its absolute maximum. No constraints are imposed on the form of P, but the ones necessary in order that P actually measures the quality of the classification process. The reject threshold is evaluated on the basis of some statistical distributions characterizing the behavior of the classifier when operating without reject option; these distributions are computed once the training phase of the net has been completed. The method has been tested with a neural classifier devised for handprinted and multifont printed characters, by using a database of about 300000 samples. Experimental results are discussed.     

Iterative learning control with advanced output data for nonlinear non-minimum phase systems

This paper investigates iterative learning control of nonlinear discrete time non-minimum phase systems in tracking problems. The main objective of this paper is to find an input-to-output mapping in order to stabilize the non-minimum phase systems and to obtain an input update law for handling uncertain systems. In conventional approaches on the tracking of non-minimum phase systems, zero dynamics is stabilized from the system equations and the input is calculated from the state information. For the learning of uncertain systems, conventional approaches depend on the output-to-state and state-to-input mappings. In the proposed method, the inverse system is stabilized using the input-to-output mapping for nonlinear non-minimum phase systems. A new input update law is proposed based on the relative degree and the number of non-minimum phase zeros. This makes the overall proposed learning system have a simple structure as in the classical ILC.     

Minimum classification error learning for sequential data in the wavelet domain

Wavelet analysis has found widespread use in signal processing and many classification tasks. Nevertheless, its use in dynamic pattern recognition have been much more restricted since most of wavelet models cannot handle variable length sequences properly. Recently, composite hidden Markov models which observe structured data in the wavelet domain were proposed to deal with this kind of sequences. In these models, hidden Markov trees account for local dynamics in a multiresolution framework, while standard hidden Markov models capture longer correlations in time. Despite these models have shown promising results in simple applications, only generative approaches have been used so far for parameter estimation. The goal of this work is to take a step forward in the development of dynamic pattern recognizers using wavelet features by introducing a new discriminative training method for this Markov models. The learning strategy relies on the minimum classification error approach and provides re-estimation formulas for fully non-tied models. Numerical experiments on phoneme recognition show important improvement over the recognition rate achieved by the same models trained using maximum likelihood estimation.     

Longitudinal modelling and control of a flapping-wing micro aerial vehicle

A simulation model, flight-dynamics oriented, of a flapping-wing micro aerial vehicle (MAV) is presented here. The model, based on animal flapping flight, integrates the aerodynamic forces computed along each wing to determine the global motion of the MAV with respect to an inertial reference frame. After some analytic simplifications, and taking into account the periodic nature of the system inputs, an averaged model is derived, and a simple, nonlinear closed-loop control law is designed for the dynamics along the vertical and pitch axis, allowing an efficient stabilization of the naturally unstable model.     

Model-based recognition of human actions by trajectory matching in phase spaces

This paper presents a human action recognition framework based on the theory of nonlinear dynamical systems. The ultimate aim of our method is to recognize actions from multi-view video. We estimate and represent human motion by means of a virtual skeleton model providing the basis for a view-invariant representation of human actions. Actions are modeled as a set of weighted dynamical systems associated to different model variables. We use time-delay embeddings on the time series resulting of the evolution of model variables along time to reconstruct phase portraits of appropriate dimensions. These phase portraits characterize the underlying dynamical systems. We propose a distance to compare trajectories within the reconstructed phase portraits. These distances are used to train SVM models for action recognition. Additionally, we propose an efficient method to learn a set of weights reflecting the discriminative power of a given model variable in a given action class. Our approach presents a good behavior on noisy data, even in cases where action sequences last just for a few frames. Experiments with marker-based and markerless motion capture data show the effectiveness of the proposed method. To the best of our knowledge, this contribution is the first to apply time-delay embeddings on data obtained from multi-view video.     

Motion recognition using nonparametric image motion models estimated from temporal and multiscale co-occurrence statistics

A new approach for motion characterization in image sequences is presented. It relies on the probabilistic modeling of temporal and scale co-occurrence distributions of local motion-related measurements directly computed over image sequences. Temporal multiscale Gibbs models allow us to handle both spatial and temporal aspects of image motion content within a unified statistical framework. Since this modeling mainly involves the scalar product between co-occurrence values and Gibbs potentials, we can formulate and address several fundamental issues: model estimation according to the ML criterion (hence, model training and learning) and motion classification. We have conducted motion recognition experiments over a large set of real image sequences comprising various motion types such as temporal texture samples, human motion examples, and rigid motion situations.     

ICA mixture models for unsupervised classification of non-Gaussian classes and automatic context switching in blind signal separation

An unsupervised classification algorithm is derived by modeling observed data as a mixture of several mutually exclusive classes that are each described by linear combinations of independent, non-Gaussian densities. The algorithm estimates the density of each class and is able to model class distributions with non-Gaussian structure. The new algorithm can improve classification accuracy compared with standard Gaussian mixture models. When applied to blind source separation in nonstationary environments, the method can switch automatically between classes, which correspond to contexts with different mixing properties. The algorithm can learn efficient codes for images containing both natural scenes and text. This method shows promise for modeling non-Gaussian structure in high-dimensional data and has many potential applications.     

Friction induced hunting limit cycles: A comparison between the LuGre and switch friction model

In this paper, friction induced limit cycles are predicted for a simple motion system consisting of a motor-driven inertia subjected to friction and a PID-controlled regulator task. The two friction models used, i.e., (i) the dynamic LuGre friction model and (ii) the static switch friction model, are compared with respect to the so-called hunting phenomenon. Analysis tools originating from the field of nonlinear dynamics will be used to investigate the friction induced limit cycles. For a varying controller gain, stable and unstable periodic solutions are computed numerically which, together with the stability analysis of the closed-loop equilibrium points, result in a bifurcation diagram. Bifurcation analysis for both friction models indicates the disappearance of the hunting behavior for controller gains larger than the gain corresponding to the cyclic fold bifurcation point.     

挖掘数据模式结构信息的混合数据分类方法

数据集中数据之间往往相互关联,所有数据整体上呈现特定的模式结构,而传统分类方法(如支持向量机)忽略数据关联信息,仅仅利用数据的物理特征(如距离、相似性等)构建数据分类模型,并在分类阶段计算测试样本与所建立分类模型间的相似性来预测测试样本的标签类型。为了解决传统分类方法利用单一数据信息的问题,提出一种挖掘数据模式结构信息的混合数据分类方法。该方法融合了两种不同类型的分类技术,将使用单一数据物理特征的传统分类方法作为普通分类方法,将利用数据模式结构信息的分类方法作为高级分类方法。特别地,该方法不仅可有效地识别数据模式结构信息以提高数据分类性能,还能提高传统分类方法的泛化能力。在人造数据集和UCI真实数据集上的大量实验结果表明了该混合数据分类方法的有效性,其分类性能优于传统分类方法。     

A new repetitive controller for mechanical manipulators

A new repetitive learning controller for motion control of mechanical manipulators undergoing periodic tasks is developed. This controller does not require exact knowledge of the manipulator dynamic structure or its parameters, and is computationally efficient. In addition, no actual joint accelerations or any matrix inversions are needed in the control law. The global asymptotic stability of the ideal and the robust stability of the nonideal control system is proven, taking into account the full nonlinear dynamics of the manipulator. Simulation results of this algorithm applied to a realistic Scara type manipulator, which includes dry friction, pay-load inertia variations, actuator/sensor noise, and unmodelled dynamics are also presented.     

On metricity of two heterogeneous measures in the presence of missing values

Heterogeneous Euclidean-overlap metric and heterogeneous value difference metric given in machine learning literature are useful for the consideration of mixed-type data for machine learning, pattern recognition and data mining tasks. Mixed-type variables are quite common in practical problems, but this property has been taken into account only seldom in pattern recognition, data mining and decision making algorithms. We observed that these two distance measures are not actually metrics after having found a special situation when they are not metric, but pseudometric, a feature to be noted while using them. Nevertheless, by changing their definitions somewhat, it is possible to meet the metricity. Especially in medical applications, the redefinition of the two measures might be important, since otherwise it is possible in theory that, for example, two identical cases would be classified differently. Nearest neighbor searching tests with medical data were run to illustrate the behavior of these measures. Notwithstanding the violation of the metricity their original forms yielded slightly better classification results. The reason was that in real data sets tested there were very few almost similar cases according to these distance measures, and the original forms based on more separating distances than the redefinitions were slightly better in the classification.